MonoER - A Edge Refined Self-Supervised Monocular Depth Estimation Method

Estimating depth from a single image is a challenging but valuable job, as it has numerous applications. Recently, some works treat depth estimating as an image synthesis problem - using stereo pairs or video sequences to train the network. In this paper, we focus on the first case - training a network with stereo image pairs. Here, we propose our MonoER (Monocular depth estimation with Edge Relining), where edge refers to the contours of the objects. Existing monocular methods often suffer from blurred edges and fail to distinguish small objects. Our network aims to solve these problems. We test our network on the KITTI dataset, showing that it produces state of the art in self-supervised monocular depth prediction. We also transfer our ideas to some popular monocular methods and the experiment results show that our ideas have the ability to generalize.